Brain network evolution trajectory prediction method, system, device and medium

Abstract

The invention discloses a brain network evolution trajectory prediction method, system and device and a medium, and relates to the technical field of artificial intelligence. Topological structure similarity and anatomical distance constraint are combined to measure the probability of connection between brain regions of a brain function network at the future moment; the constraint effect of the dissection distance constraint is quantified through a power law attenuation model so as to conform to the economical principle of brain network connection, the topological structure similarity calculated by nodes on mutual information of inter-node multi-order paths can quantify the synergistic effect of multi-order topological structures in the brain network, the capture ability of the model on high-order pathological features is improved, and the brain network connection efficiency is improved. Therefore, the dynamic evolution trajectory and process of the brain network connection from the healthy person to the patient are clearly simulated based on the connection probability, and more reasonable assistance is provided for analyzing the pathological mechanism of depression from the brain network level.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system, device, and medium for predicting the evolutionary trajectory of brain networks. Background Technology

[0002] Depression, a common mental disorder, affects people's physical and mental health, social function, and quality of life. The brain is a complex network composed of different brain regions connected by structural and functional connections. During the development of depression, the brain exhibits functional abnormalities and compensatory phenomena, which are reflected at the brain network level, specifically manifested in the establishment and deletion of functional connections. Generally, brain networks exhibit significant dynamic evolutionary characteristics in both physiological activities and pathological processes. Under physiological conditions, brain network evolution manifests as functional connection reorganization and connection weight optimization, optimizing information transmission efficiency through real-time regulation of network state, thereby completing higher cognitive functions such as memory and decision-making. Under pathological conditions, brain network evolution manifests as degenerative changes in structural connections or dysfunctional compensatory changes in functional connections, leading to disordered neural information processing, and consequently causing functional impairments in cognition, emotion, and behavior. Therefore, exploring the dynamic evolutionary laws of brain networks is of great research significance for revealing the physiological and pathological mechanisms of depression.

[0003] Currently, research on the dynamic evolution of brain networks generally focuses on lateral evolution. Lateral evolution refers to the construction of cross-population common network evolution models under the constraints of pathological mechanisms, based on cross-sectional data (i.e., data collected from multiple groups of brain networks at the same time). This aims to uncover the shared topological reconstruction patterns of the population and reveal the dynamic transformation of brain networks from normal to pathological states. Subsequently, by analyzing the cross-population common evolutionary trajectory, the topological preferences of cross-brain region information transmission paths can be revealed, and the dynamic decay patterns of early biomarkers can be analyzed, which helps to understand disease mechanisms from a dynamic perspective. The most widely used lateral evolution prediction framework is based on the principle of brain economy and uses link prediction methods to calculate the connection probability between brain regions. Because this framework directly relates to biological mechanisms by simulating the disconnection and compensation phenomena of brain connections, it has good interpretability.

[0004] Current link prediction methods mainly include similarity-based methods, probabilistic models, and machine learning-based methods. However, the evolution of brain networks in pathological processes involves not only changes in direct connections between nodes (first-order neighbors) but also the synergistic effects of multi-order local topologies (such as second-order and third-order paths). Current link prediction methods mainly rely on the low-order topological properties of nodes during evolutionary modeling, often neglecting the effective modeling of the synergistic effects of multi-order local topologies, making it difficult to predict the explicit dynamic evolutionary patterns of brain networks. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for predicting the evolutionary trajectory of brain networks, which can solve the problem in the prior art that it is difficult to predict the clear dynamic evolution law of brain networks.

[0006] This invention provides a method for predicting the evolutionary trajectory of brain networks, comprising the following steps: Obtain the brainwave signals of healthy individuals and those of patients with depression. Using the brain regions corresponding to each channel in the EEG signals of healthy individuals and patients with depression as nodes, and the brain regions transmitting functional EEG signals through the corresponding channels, a graph network is constructed to form a brain functional network that includes both healthy individuals and patients by using the connections between the brain regions corresponding to the channels as edges. Brain regions associated with depression were extracted from the brain functional network to form brain region subnetworks; in these subnetworks, corresponding nodes from the brain regions of healthy individuals and patients with depression were paired. For multiple node pairs contained in the brain functional network, the anatomical distance between nodes within the node pair is obtained, and the anatomical distance is subject to exponential decay constraint. There are multiple transmission paths between nodes within the node pair to transmit information. The mutual information of the multiple transmission paths between nodes within the node pair is obtained, and the topological similarity between nodes within the node pair is obtained based on the mutual information. The product of the constrained anatomical distance and the topological similarity is set as the connection probability between the corresponding nodes within the node pair, and the set of connection probabilities of all node pairs in the brain region subnetwork is obtained. Using the brain functional network of healthy individuals as the starting point of evolution and the brain functional network of patients as the ending point of evolution, the evolution rules from the brain functional network of healthy individuals to the brain functional network of patients are set: based on the edge difference rate between the edges in the brain functional network of healthy individuals and the edges in the brain functional network of patients, edges are added and deleted in the brain functional network of healthy individuals. When adding edges, the edge connection is established between the nodes with the highest connection probability in the connection probability set from the unconnected node pairs. When deleting edges, the edge connection is broken between the nodes with the lowest connection probability in the connection probability set from the already connected node pairs. The brain functional network of healthy individuals is evolved into the brain functional network of patients according to the evolutionary rules, resulting in the final simulated brain network. The trajectory of the edges in the brain functional network of healthy individuals to the edges in the final simulated brain network is extracted from the evolutionary process, so as to predict the evolution trajectory of the brain network.

[0007] This invention also provides a brain network evolution trajectory prediction system, comprising: The information acquisition module is used to acquire brainwave signals from healthy individuals and patients with depression. Using the brain regions corresponding to each channel in the EEG signals of healthy individuals and patients with depression as nodes, and the brain regions transmitting functional EEG signals through the corresponding channels, a graph network is constructed to form a brain functional network that includes both healthy individuals and patients by using the connections between the brain regions corresponding to the channels as edges. Brain regions associated with depression were extracted from the brain functional network to form brain region subnetworks; in these subnetworks, corresponding nodes from the brain regions of healthy individuals and patients with depression were paired. The connectivity probability modeling module is used to obtain the anatomical distance between nodes within a pair of nodes in a brain functional network, and to impose an exponential decay constraint on the anatomical distance. Since there are multiple transmission paths between nodes within a pair to transfer information, the module obtains the mutual information of these multiple paths, and then obtains the topological similarity between nodes within the pair based on the mutual information. The product of the constrained anatomical distance and the topological similarity is set as the connectivity probability between nodes within the corresponding pair, and the module also obtains the connectivity probability set of all node pairs in the brain region subnetwork. The evolution module is used to set the evolution rules from the healthy brain functional network to the patient brain functional network, with the healthy brain functional network as the starting point and the patient brain functional network as the ending point. Based on the edge difference rate between the edges in the healthy brain functional network and the edges in the patient brain functional network, edges are added and deleted in the healthy brain functional network. When adding edges, the edge connection is established between the nodes with the highest connection probability in the connection probability set from the unconnected node pairs. When deleting edges, the edge connection is broken between the nodes with the lowest connection probability in the connection probability set from the connected node pairs. The brain functional network of healthy individuals is evolved into the brain functional network of patients according to the evolutionary rules, resulting in the final simulated brain network. The trajectory of the edges in the brain functional network of healthy individuals to the edges in the final simulated brain network is extracted from the evolutionary process, so as to predict the evolution trajectory of the brain network.

[0008] This invention also provides an electronic device, including a memory and a processor; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the brain network evolution trajectory prediction method as described above.

[0009] This invention also provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of a brain network evolution trajectory prediction method as described above.

[0010] This invention provides a method, system, device, and medium for predicting the evolutionary trajectory of brain networks. Compared with the prior art, its advantages are as follows: This invention combines topological similarity and anatomical distance constraints to measure the connection probability between nodes within a brain functional network at future time points. The anatomical distance constraint is quantified using a power-law decay model to ensure the economic efficiency of brain network connections. The topological similarity calculated from the mutual information of multi-order paths between nodes within a network can quantify the synergistic effect of multi-order topological structures in the brain network, improving the model's ability to capture higher-order pathological features. Thus, based on connection probability, it can clearly simulate the dynamic evolution trajectory and process of brain network connections from healthy individuals to patients, providing a more reasonable aid for analyzing the pathological mechanisms of depression at the brain network level. Attached Figure Description

[0011] Figure 1 This is a schematic diagram illustrating the implementation process of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; Figure 2 A schematic diagram of the original electroencephalogram (EEG) signal segments of a patient with depression and a healthy subject in an embodiment of the present invention for predicting the evolution trajectory of a brain network; wherein: (a) is a patient with depression; (b) is a healthy subject; Figure 3 This is a schematic diagram of EEG signal segment preprocessing for a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a) is before preprocessing; (b) is after preprocessing; Figure 4 This is a schematic diagram of the functional connectivity matrix of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a), (b) and (c) show the functional connectivity matrix diagram of three EEG segments of a healthy subject 02020019; (d), (e) and (f) show the functional connectivity matrix diagram of three EEG segments of a patient with depression 0210030; Figure 5 This is a schematic diagram of the adjacency matrix of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a), (b) and (c) show the adjacency matrix diagram of three EEG segments of a healthy subject 02020019; (d), (e) and (f) show the adjacency matrix diagram of three EEG segments of a patient with depression 0210030; Figure 6 Violin plots of global indices of whole-brain functional networks for the MDD and NC groups of a brain network evolution trajectory prediction method provided in this embodiment of the invention; wherein: (a) is local efficiency; (b) is average clustering coefficient; (c) is average shortest path length; (d) is global efficiency; Figure 7 This is a schematic diagram of the location distribution of disease-related brain regions in a brain network evolution trajectory prediction method provided by an embodiment of the present invention; Figure 8This is a schematic diagram of the hub node location distribution of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a) is a hub node location distribution diagram of the NC group; (b) is a hub node location distribution diagram of the MDD group; Figure 9 This is a schematic diagram showing the location distribution of key brain regions in a brain network evolution trajectory prediction method provided by an embodiment of the present invention; Figure 10 A schematic diagram of the edge difference rate change curve of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; Figure 11 This is a schematic diagram of the evolution process of key brain region subnetworks in a subject according to a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a) is the initial brain network; (b), (c), (d) and (e) are simulated brain networks in the evolution process of healthy individuals; and (f) is the final simulated brain network; Figure 12 This is a schematic diagram of the high-frequency changes in functional connections during the evolution process of a brain network evolution trajectory prediction method provided in an embodiment of the present invention; wherein: (a) represents the functional connection that is broken most frequently; and (b) represents the functional connection that is established most frequently. Detailed Implementation

[0012] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0013] See Figure 1 This invention provides a method for predicting the evolutionary trajectory of brain networks. By focusing on brain regions related to the pathology of depression, this invention selects key brain regions, constructs subnetworks of these key regions, highlights pathology-related dynamic characteristics, and effectively avoids redundant connections at the whole-brain scale. This allows for a more accurate characterization of disease-related network reorganization features, improves the accuracy of pathological mechanism analysis, and reduces computational complexity. Furthermore, it models connection probabilities using topological similarity and anatomical distance constraints. The anatomical distance constraint is quantified using a power-law decay model to ensure economic efficiency in brain network connectivity. Multi-order local mutual information (MOMI) quantifies the contribution of paths of different lengths to topological similarity, thus achieving an effective representation of the multi-order topological dynamic characteristics of brain networks. Through the synergistic effect of these two methods, a more reasonable brain network evolution model is constructed to simulate the dynamic transformation process of the brain network state from a healthy control group to a patient with severe depression, providing a quantitative basis for the analysis of potential connection mechanisms of depression at the brain network level.

[0014] Specifically, it includes the following steps: Step S1: Preprocess the electroencephalogram (EEG) signals of the patients with depression (MDD group) and the normal control group (NC group). Specifically, this includes:

[0015] For a given set of EEG signals ,in The number of signals in the MDD group. The number of signals in the NC group; for the signal set Preprocessing is performed, including interpolation bad derivatives, bandpass filtering, and artifact removal.

[0016] Step S2: Construct a whole-brain functional network from the preprocessed EEG signals. This specifically includes:

[0017] Step S21: For the preprocessed signal, the phase-locking value (PLV) is used to measure the correlation between brain region signals, and a functional connectivity matrix is ​​constructed. Specifically:

[0018] For the preprocessed signal ,in Number of EEG signal channels Given the signal length, the functional connection between each pair of channels is calculated using the phase-locked loop (PLL) value. First, the functional connection between each pair of channels is calculated, denoted as... The functional connectivity matrix constructed accordingly is as follows:

[0019] .

[0020] Step S22: Further set the connection threshold This is used to binarize the aforementioned functional connection matrix to obtain the adjacency matrix. Specifically:

[0021] The functional connectivity matrix is ​​binarized and represented as follows: The resulting matrix after processing is an adjacency matrix, represented as: Step S23: Using the brain region corresponding to each channel in the EEG signal as a "node," determine whether there is a connection between the corresponding brain regions based on the adjacency matrix, i.e., whether there is an "edge," thereby constructing a "graph" of the corresponding signals. Specifically:

[0022] Using the brain region corresponding to each channel in the EEG signal as a "node", when the adjacency matrix At that time, determine the channel and There are connections between corresponding brain regions, i.e., there are "edges"; based on this, corresponding signals can be constructed. The “diagram” is denoted as .

[0023] in, Represents a set of nodes; Represented as the number of nodes; Represents an edge set; This is expressed as the number of nodes.

[0024] The result is For signal The whole brain functional network, memory For signal set A collection of whole-brain functional networks.

[0025] Step S3: Design an algorithm to screen key brain regions associated with brain diseases and construct subnetworks of these key brain regions. This specifically includes:

[0026] Step S31: Construct an EEG signal set according to the brain functional network construction method in step S2. Brain networks containing all EEG signals .

[0027] Step S32: For each brain network compute nodes Local index value (node ​​degree) Clustering coefficient Local efficiency and betweenness centrality ).

[0028] Step S33: Perform the Mann-Whitney U test on each local index of each node to verify whether there are statistically significant differences in the local characteristics between the MDD group and the NC group.

[0029] Step S34: For node If the hypothesis test results for all local indicators show no significant difference between the MDD group and the NC group, then it is considered that... This node is not related to the disease-related brain region; delete it.

[0030] Step S35: Average the functional connectivity matrices of the subjects within each group to obtain the group network. and calculate the nodes Proximity centrality of nodes in a group network The proximity centrality are respectively .

[0031] Step S36: If the nodes in the group network If the proximity centrality is one standard deviation higher than the average proximity centrality of the group network, then it is considered... The node is a hub node.

[0032] Step S37: The set of disease-related nodes and pivotal nodes is called the "critical brain region," denoted as... .

[0033] Step S38: Constructing a set of EEG signals Key brain regions and subnetworks. Specifically:

[0034] First of all, for Signals in Key brain regions were calculated using a key brain region screening algorithm. And extract the signal from the corresponding channel, denoted as Secondly, regarding signals A brain functional network is constructed using the method in step S2, denoted as . ,say For signal Key brain subnetworks; finally, for signal sets All signals are analyzed, and their key brain region subnetworks are constructed to obtain the signal set. The key brain region subnetwork set, denoted as .

[0035] Step S4: Combining topological structure and anatomical distance factors, construct a brain network connectivity probability model based on multi-order local mutual information. Specifically, this includes:

[0036] Given a pair of nodes The connection probability between them can be expressed as: in: Represents a node and nodes Anatomical distance between; anatomical distance constraint is defined as To reflect the economic principle of brain connectivity, namely the biological characteristics of neurons optimizing connection layout under the constraints of spatial proximity and metabolic cost; the constraint of anatomical distance is quantified by a power-law decay model, which can simultaneously characterize the local density of short-range connections and the preservation mechanism of key long-range connections.

[0037] Represents a node and The topological similarity between nodes is calculated based on mutual information between nodes using multi-order paths. Studies have shown that mutual information can quantify nonlinear associations between nodes and reveal the information flow between brain regions through indirect paths, which is crucial for understanding the connection patterns of brain diseases. Higher mutual information indicates that the uncertainty of establishing a connection between two nodes in the future is greatly reduced.

[0038] In link prediction, the connection probability of unconnected node pairs is positively correlated with their topological similarity; considering nodes u and v, It is a node and The length between is The set of all paths, the conditional entropy can be expressed as: in: Represents a node and nodes There are connections between them. node and nodes There is no connection between them; It is the maximum path length we consider in brain networks.

[0039] For two unconnected nodes, given existing path conditions, the smaller the conditional entropy, the more likely the node is to be connected. and The probability of connection between them is relatively high, i.e., topological similarity. The topological similarity index, based on multi-order path mutual information, is defined as the negative of the conditional entropy, expressed as: in: It is a length of The weight of the path, ; , , These are nodes , and The degree.

[0040] For any two nodes and Calculate the node using the above formula and The probability of connections between them.

[0041] Step S5: Design evolutionary rules based on connectivity probabilities, simulate disease progression through iterative prediction, and obtain the final simulated brain network of the evolutionary stages. Specifically, this includes:

[0042] Step S51: Network initialization, obtaining the initial brain network group and the target brain network group, calculating multiple global network indices of the target brain network group, and defining the edge sets of the initial brain network group and the target brain network group. Specifically:

[0043] For EEG signal collections Through the method in step S3, the key brain region subnetwork set is as follows: The key brain region subnetworks of the NC group are denoted as the initial brain network group. The key brain region subnetworks of MDD are denoted as the target brain network group. ,in , These represent the number of subjects in the initial brain network group and the target brain network group, respectively.

[0044] Computational target brain network Average shortest path length Average clustering coefficient Local efficiency and overall efficiency The edge sets of the initial brain network group and the target brain network group are defined as follows: and The number of corresponding edge sets is denoted as and The mean number of edge sets is denoted as and .

[0045] Step S52: Design an evolutionary decision based on the difference in the number of edges to select whether to add or delete functional connections. Specifically:

[0046] The relative difference between the current number of edges and the target number of edges is defined as the edge difference rate, expressed as: .

[0047] The increasing probability function is defined as: in: This represents a difference-oriented adjustment factor; if the current number of edges is less than the target number of edges, ,at this time Furthermore, the greater the difference, the more inclined the edge will be to increase. If the difference between the current number of edges and the target number of edges is small, there is a 50% probability that the edge increase operation will be selected.

[0048] This edge-increasing function enables rapid adjustment when there are significant differences in the number of edges, and fine-tuning when approaching the target; it also generates random numbers. ,like If the condition is met, an edge addition operation is performed; otherwise, an edge deletion operation is performed. By introducing random factors, the external environmental interference experienced by the brain is simulated on the one hand, and the problem of getting trapped in local optima is avoided on the other.

[0049] Step S53: Calculate the simulated brain network using the brain network connectivity probability model described in step S4. The connection probability between any two nodes in the equation. The number of evolution iterations is sorted according to the magnitude of the connection probability.

[0050] Step S54: Determine the first The number of edges participating in the evolution, defined as the adaptive step size. Among them, based on network size This represents the maximum number of edges in a single operation.

[0051] Step S55: Perform an evolutionary operation to update the edges in the simulated brain network, obtaining the current simulated brain network. Specifically:

[0052] right The edges in the array are updated. If an addition operation is performed, a node pair without an existing edge is selected from the set of nodes with the highest-ranked connection probabilities to establish a connection. If a deletion operation is performed, then select a pair of nodes with known connections from the lowest-ranked set of minimum connection probabilities to disconnect them. There are several connections; it is important to note that when performing a deletion operation, at least one connection must be retained for each edge to ensure the connectivity of the simulated network, thus obtaining the current simulated brain network. .

[0053] Step S56: Calculate the number of edges in the current simulated brain network group. Number of edges connected to the target brain network group The p-value of the two-sample t-test.

[0054] Step S57: When the number of edges in the two groups of brain networks is compared using a two-sample t-test... Or reach the maximum number of evolutions The entire evolutionary process has ended, and the currently obtained simulated brain network will be regarded as the final simulated brain network of the evolutionary stage. .

[0055] Step S6: Evaluate the performance of the proposed brain network evolution model. Perform Mann-Whitney U tests on multiple global network metrics of the simulated brain network and the target brain network, define the evolutionary loss function, and achieve evolutionary evaluation of the simulated brain network. Specifically, this includes:

[0056] Simulated brain networks Target brain network Mann-Whitney U test was performed on multiple global network metrics to calculate the simulated brain network genome. Average shortest path length Average clustering coefficient Local efficiency and overall efficiency This paper analyzes the overall function and characteristics of a network from four perspectives: distance characteristics of information transmission, connection density, local information transmission efficiency, parallel capability, and integration. An evolutionary loss function is defined. This enables the evaluation of the evolution of simulated brain networks; its evolutionary loss function... Represented as: in: , , , The p-value represents the Mann-Whitney U test result for the average shortest path length, average clustering coefficient, local efficiency, and global efficiency.

[0057] The smaller the value, the stronger the simulated network generated by the evolutionary model. Network properties and the target brain network of real MDD groups More similar.

[0058] Specific experiments include: Step S1: Preprocess the EEG signals of the patients with depression (MDD group) and the normal control group (NC group), specifically including: Using the MODMA dataset (http: / / modma.lzu.edu.cn / data / index / ), 24 patients with depression and 29 healthy controls admitted to the Department of Psychiatry of the Second Hospital of Lanzhou University were included. After strict matching, there were no statistically significant differences in age and gender distribution between the two groups. All participants signed informed consent forms, and the severity of depressive symptoms was assessed using the PHQ-9 scale, while anxiety levels were quantified using the GAD-7 scale. The study protocol was approved by the hospital's medical ethics committee. The experiment used the EGI (Electrical Geodesics Inc.) 128-channel HydroCel system to record resting-state EEG data with eyes closed at a sampling rate of 250 Hz, with the Cz electrode used as the reference electrode. Before data acquisition, it was confirmed that the impedance of each electrode was below 50 kΩ, and the experiment was conducted in a quiet environment with good electromagnetic shielding. Subjects were required to keep their heads still and avoid blinking and eye movements as much as possible, completing a 5-minute baseline recording while fully awake with eyes closed.

[0059] Raw EEG signal fragments from patients with depression and healthy subjects, such as Figure 2 As shown, the MNE toolkit is used to perform standardized preprocessing on the raw EEG data; the specific process is as follows: Data filtering: First, power frequency interference is eliminated by using a 50 Hz notch filter, and then effective frequency band signals are screened using an FIR bandpass filter from 4 Hz to 30 Hz. Interpolation bad conduction: Linear interpolation is performed to repair the detected defective electrode channels to ensure signal continuity; Whole-brain average reference: The average signal of all electrodes in the whole brain is used as a new reference point, and the influence of the reference electrode on subsequent analysis is eliminated by the projection method. Independent Component Analysis (ICA): The FastICA algorithm was used to separate electrooculography (EOG) artifacts from electromyography (EMG) artifacts. Signal segmentation and bad segment removal: The continuous EEG recording was divided into non-overlapping segments of 5 seconds in length, and abnormal segments with amplitudes exceeding ±150 were removed. To ensure that the number of segments was consistent among the subjects, this invention selected 16 consecutive and valid EEG data segments for each subject for subsequent analysis, with a total data duration of 80 seconds.

[0060] Subject 0210026's EEG signal fragments before and after pretreatment are as follows: Figure 3 As shown; after removing the reference electrode Cz, there are a total of 128 electrode channels. The 128 electrodes are divided into 9 brain regions according to their distribution location, including the left frontal lobe, right frontal lobe, left temporal lobe, right temporal lobe, left central region, right central region, left parieto-occipital region, right parieto-occipital region, and central region.

[0061] Step S2: Construct a whole-brain functional network with a functional connectivity matrix of 128×128 dimensions. Figure 4 (a)-(c) show color-coded functional connectivity matrices of three EEG segments from healthy subject 02020019. Figure 4 (d)-(f) show the color map of the functional connectivity matrix of three EEG segments of patient 0210030 with depression.

[0062] The functional connectivity matrix is ​​binarized. Based on network sparsity analysis, the binarization threshold is set to 0.5 to preserve significant functional connections, thus obtaining the corresponding adjacency matrix. Figure 5 (a)-(c) show the adjacency matrix color plots of three EEG segments from the healthy subject 02020019 mentioned above. Figure 5 (d)-(f) show the adjacency matrix color maps of the three EEG segments corresponding to the aforementioned patient 0210030 with depression; each pixel in the map reflects whether there is a connection between the corresponding two channels, with purple representing a connection and blue representing no connection. From Figure 4As shown in Figure 5, there are significant differences in the topological structures of the functional connectivity matrix and adjacency matrix between patients with depression and healthy subjects. The color modules of healthy subjects are clearly defined, reflecting a moderate separation between functional networks. Patients with depression have lower functional connectivity strength, abnormal connections in local brain regions, and cross-module functional compensation phenomena.

[0063] For the whole-brain functional networks of patients with depression and healthy subjects, this invention further calculated four global indicators (including average clustering coefficient, average shortest path length, local efficiency, and global efficiency). For ease of subsequent description, these are referred to as "MDD group global indicators" and "NC group global indicators," respectively. Then, the Mann-Whitney U test was used to determine whether there were significant differences in the global indicators.

[0064] Figure 6 shows violin plots of four global indicators for the MDD and NC groups. The violin shape represents the density plot, which reflects the concentration of data distribution; the wider the plot, the more concentrated the data distribution. The figure shows that the average clustering coefficient of the NC group is more concentrated than that of the MDD group, although there is a difference in their medians. The average shortest path length of the MDD group differs significantly from that of the NC group. The global efficiency of the MDD group is more uniformly distributed than that of the NC group, but there is a significant difference in their medians. The median global efficiency of the NC group is higher than that of the MDD group, indicating reduced information transmission efficiency in the brain networks of MDD patients.

[0065] The results showed that the Mann-Whitney U tests for all four global indicators were significantly different between the MDD and NC groups. The network characteristics presented in this invention are consistent with physiological experimental findings. Mulders et al. pointed out that white matter microstructural damage in depressed patients increases communication costs between nodes, resulting in decreased global efficiency and increased average shortest path length, indicating impaired network integration capabilities. Dai et al. indicated that the average clustering coefficient of untreated MDD patients was higher than that of NC patients, and the clustering coefficient of anterior cingulate cortex nodes was positively correlated with depression severity. Buch et al. pointed out that changes in local efficiency in the prefrontal cortex and limbic system are associated with emotion regulation disorders.

[0066] Step S3: Using a key brain region screening algorithm, a total of 57 disease-related key brain regions were identified. Figure 7 The study revealed key brain regions associated with the disease, including the parieto-occipital region, temporal lobe, and central region. Brain regions confirmed to be associated with depression showed significantly more abnormal regions in the right hemisphere than in the left. Abnormal functional connectivity in the right hemisphere is closely related to mood regulation disorders in depression, and the experimental results of this invention further support this hypothesis. The distribution of pivot nodes in the MDD and NC groups is shown below. Figure 8 As shown; the final set of key brain regions containing 67 brain regions was obtained as follows. Figure 9 As shown, key brain region subnetworks were constructed based on this.

[0067] The experimental results of this invention are consistent with previous physiological experimental results, which showed that the right hemisphere of patients with depression has more significant functional deficits compared to the left hemisphere; the frontal lobe has been identified as one of the brain regions closely related to MDD, and the frontal lobe of MDD patients exhibits an asymmetrical connectivity pattern; the study by Menon et al. proposed that a decrease in the clustering coefficient (CC) of the prefrontal cortex, striatum, and medial temporal cortex may trigger depressive symptoms, such as persistent sadness, feelings of worthlessness, and recurring negative thinking; Fan et al. found that abnormal activity in the right superior temporal gyrus may be a potential biomarker for suicidal tendencies in MDD patients; in addition, Blackhart et al. believed that activity in the right parietal-temporal lobe is closely related to the severity of depressive symptoms.

[0068] Step S4: In the implementation example, due to the local connectivity characteristics of brain networks, the contribution of node connection probability decays as path length increases. Therefore, the longest path length is considered. .

[0069] Step S5: Design evolutionary rules based on connection probability, simulate disease progression through iterative prediction, and obtain the final simulated brain network of the evolutionary stage.

[0070] In the experimental example, a brain network evolution method based on distance constraints and multi-order local mutual information (DSMOMI) was used to simulate the evolution from a healthy brain network to a brain network with depression; the initial temperature of the simulated annealing algorithm was also used. Cooling coefficient Maximum number of iterations The hyperparameters are solved using the simulated annealing algorithm. , , ,because Then the corresponding At this point, the similarity evaluation function between the generated simulated brain network and the target brain network... Minimum.

[0071] Differential-oriented regulatory factors for parameter selection during evolution. Maximum evolution .

[0072] like Figure 10As shown, the trend of the edge difference rate between the simulated brain network and the target brain network with the number of iterations is presented during the dynamic evolution process. In the initial evolution stage, the topological differences between the simulated brain network and the target brain network are large, and rapid network evolution is achieved through the dynamic adjustment mechanism of adaptive step size. When the evolution enters the middle stage, the decreasing trend of the difference rate slows down, indicating that the network topology has entered the fine adjustment stage. In the later stage of evolution, the number of network edges reaches a steady-state equilibrium, and the fluctuation of the difference rate does not exceed 2%. It maintains efficient convergence in the global search stage and numerical stability in the local optimization stage.

[0073] like Figure 11 As shown, the evolutionary trajectory of the sub-networks of key brain regions in the subjects is illustrated; (a) is the initial brain network, (b)-(e) are the simulated brain networks during the evolution of healthy individuals, and (f) is the final simulated brain network. The initial brain network shows symmetrical distribution of connections between the right and left hemispheres without significant lateralization, consistent with the functional separation-integration balance of healthy brain networks, and exhibits high global efficiency. During the evolution, functional connectivity in the right temporal lobe region is significantly weakened, network connections become sparse, cross-hemispheric connections decrease, and there is a trend of network splitting into isolated sub-modules.

[0074] Step S6: Evaluate the performance of the proposed brain network evolution model, perform Mann-Whitney U tests on multiple global network indices of the simulated brain network and the target brain network, define the evolution loss function, and realize the evolution evaluation of the simulated brain network.

[0075] To further verify the effectiveness of the proposed method, the values ​​of four network attributes and similarity evaluation functions of the simulated brain network and the target brain network obtained from different evolutionary models were compared. ; The results show that there are no statistically significant differences in the network properties between the simulated brain network and the target brain network. The smaller the difference, the better the evolutionary model can simulate the brain network evolution process. The comparison results are shown in Table 1, and the key brain region evolution validity is verified in Table 2.

[0076] Table 1 Comparison of the evolutionary performance of link prediction methods Table 1 compares the values ​​of four network attributes and similarity evaluation functions of the simulated brain network and the target brain network obtained from different evolutionary models. In Table 1, CN, AA, PA, RA, JC, LNB, MI, and LP represent the methods used to quantify topological similarity and calculate the probability of edges between nodes using different link prediction methods. "DS" indicates that anatomical distance constraints are included when calculating the probability of edges between nodes. Rand indicates that edge probability calculation is not required, and the added or deleted edges are randomly selected. All methods maintain consistency in their evolutionary steps, except for differences in edge probability calculation.

[0077] Based on the performance comparison of link prediction methods in Table 1, the DSMOMI model proposed in this invention demonstrates significant advantages in simulating brain networks associated with depression. Specifically, the simulated brain network generated by DSMOMI exhibits superior performance in terms of average shortest path length (…). ), average clustering coefficient ( ), local efficiency ( ) and global efficiency ( No significant differences were found in the statistical tests of indicators such as ) This indicates that its network topology properties are statistically consistent with the target depression brain network. From the perspective of comprehensive similarity assessment indicators, the DSMOMI value of 0.7805 is significantly lower than other models, further validating the higher global similarity between its simulated network and the target network. Although the DSJC model achieves a large value in local efficiency, its average shortest path length (...) ) and average clustering coefficient ( The simulation performance of MOMI is significantly worse than that of DSMOMI, failing to achieve good evolutionary performance across multiple attributes. Compared to traditional models based on common neighbors (CN, AA, etc.) and local topological indices (JC, LNB, etc.), MOMI, by quantifying the inter-node information interaction of multi-order path lengths, can distinguish the differences in the contribution of different path lengths to functional connectivity, avoiding the homogenization of multi-order paths by LP methods, and thus more accurately capturing the hierarchical characteristics of brain network evolution.

[0078] Table 2 Validation of the Evolutionary Effectiveness of Key Brain Regions As shown in Table 2, the evolutionary modeling strategy based on key brain regions is significantly superior to the whole-brain network analysis method DSMOMI (Key Brain Regions). The SI value of DSMOMI (Whole Brain) is 8.3% lower than that of DSMOMI (Whole Brain). Key brain region selection, by focusing on brain regions related to the pathology of depression, effectively avoids redundant connection interference at the whole-brain scale, thus more accurately characterizing the network reorganization features related to the disease. In addition, the synergistic mechanism of integrating topological similarity and anatomical distance constraints plays a decisive role in improving evolutionary accuracy. Experimental data show that relying solely on topological similarity (… ) or simply apply distance constraints ( The performance of the model is significantly worse than that of the combined DSMOMI model. Anatomical distance constraints suppress the generation of long-range connections through an exponential decay function, which aligns with the energy optimization principle of brain network connections. Multi-order local mutual information (MOMI) quantifies the contribution of paths of different lengths to topological similarity, thereby achieving higher network efficiency. The synergistic effect of these two mechanisms enables brain network evolution.

[0079] like Figure 12(a) The 20 most frequently disconnected connections are concentrated in the right parietal-occipital and right central regions, which are closely related to the functional decoupling of the right hemisphere-dominated emotional processing network in patients with depression, supporting the theory that right hemisphere functional impairment is dominant in depression. Figure 12 (b) shows that newly established high-frequency connections are concentrated in the left frontal lobe and left central region, reflecting that brain networks achieve functional compensation through cross-hemispheric reorganization. Figure 12 Visualizing the functional abnormality-compensation mechanism reveals the core mechanism of dynamic imbalance in the brain network of depression, providing a potential biomarker for further understanding the neural mechanisms of depression.

[0080] This invention designs a method for constructing key brain region subnetworks based on topological indices of the whole-brain functional networks of patients with brain diseases and healthy subjects. Furthermore, it combines topological similarity and anatomical distance to measure the probability of brain network connectivity at future time points, and incorporates an evolutionary strategy to design a brain network lateral evolution prediction method based on multi-order local mutual information to simulate the evolution of brain network connectivity patterns during disease progression. This method can focus on disease-related brain regions and quantify the synergistic effects of multi-order topological structures in the brain network, improving the model's ability to capture high-order pathological features while reducing computational complexity. Considering the selective damage mechanism of diseases to specific brain regions, a "key brain region" screening algorithm is designed to construct key brain region subnetworks for subsequent simulation of brain network evolution during disease progression. This reduces computational complexity while more accurately revealing changes in functional connectivity between brain regions affected by depression. Addressing the insufficient representation of multi-order topological features in lateral evolution prediction, a connectivity probability measurement method is designed using multi-order local mutual information, and an evolutionary strategy best suited to the task requirements is proposed, improving the overall system performance and enhancing model evolution performance. This provides a more accurate and reasonable basis for analyzing the pathological mechanisms of depression from the brain network level.

[0081] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for predicting the evolutionary trajectory of brain networks, characterized in that, Includes the following steps: Obtain the brainwave signals of healthy individuals and those of patients with depression. Using the brain regions corresponding to each channel in the EEG signals of healthy individuals and patients with depression as nodes, and the brain regions transmitting functional EEG signals through the corresponding channels, a graph network is constructed to form a brain functional network that includes both healthy individuals and patients by using the connections between the brain regions corresponding to the channels as edges. Brain regions associated with depression were extracted from the brain functional network to form brain region subnetworks; in these subnetworks, corresponding nodes from the brain regions of healthy individuals and patients with depression were paired. For multiple node pairs contained in the brain functional network, the anatomical distance between nodes within the node pair is obtained, and the anatomical distance is subject to exponential decay constraint. There are multiple transmission paths between nodes within the node pair to transmit information. The mutual information of the multiple transmission paths between nodes within the node pair is obtained, and the topological similarity between nodes within the node pair is obtained based on the mutual information. The product of the constrained anatomical distance and the topological similarity is set as the connection probability between the corresponding nodes within the node pair, and the set of connection probabilities of all node pairs in the brain region subnetwork is obtained. Using the brain functional network of healthy individuals as the starting point of evolution and the brain functional network of patients as the ending point of evolution, the evolution rules from the brain functional network of healthy individuals to the brain functional network of patients are set: based on the edge difference rate between the edges in the brain functional network of healthy individuals and the edges in the brain functional network of patients, edges are added and deleted in the brain functional network of healthy individuals. When adding edges, the edge connection is established between the nodes with the highest connection probability in the connection probability set from the unconnected node pairs. When deleting edges, the edge connection is broken between the nodes with the lowest connection probability in the connection probability set from the already connected node pairs. The brain functional network of healthy individuals is evolved into the brain functional network of patients according to the evolutionary rules, resulting in the final simulated brain network. The trajectory of the edges in the brain functional network of healthy individuals to the edges in the final simulated brain network is extracted from the evolutionary process, so as to predict the evolution trajectory of the brain network.

2. The brain network evolution trajectory prediction method according to claim 1, characterized in that, The acquisition of the brain functional network includes: The electroencephalogram (EEG) signals of the healthy individuals and patients with depression were as follows: ； in: Indicates the number of EEG signal channels; Indicates the signal length; This indicates the number of signals in the patient's electroencephalogram (EEG). The number of signals representing the brain electrical signals of a healthy person; Obtaining the functional connectivity between two channels in an EEG signal Obtaining EEG signals based on functional connectivity The functional connection matrix is ​​represented as follows: ； Preset connection threshold The functional connectivity matrix is ​​binarized to obtain the adjacency matrix. , represented as: ； Using the brain region corresponding to each channel in the EEG signal as a "node", when the adjacency matrix At that time, determine the channel and There are connections between corresponding brain regions, that is, there are "edges", which can construct corresponding signals. Whole brain functional network ,in For a set of nodes; Indicates the number of nodes; It is an edge set; Indicates the number of sides.

3. The brain network evolution trajectory prediction method according to claim 2, characterized in that, The acquisition of the brain region subnetwork includes: Acquiring brain functional networks internal nodes node degree Clustering coefficient Local efficiency and betweenness centrality For each node, a Mann-Whitney U test was performed on its local indicators. If the test results showed no statistically significant difference in local indicators between healthy individuals and patients, then the results were considered... Delete the corresponding node if it is not a disease-related brain region. The functional connectivity matrices of subjects within both the patient and healthy groups were averaged to obtain the group brain network. and calculate the nodes Proximity centrality of nodes in a group network The proximity centrality are respectively If the nodes in the group network If the proximity centrality of a node is one standard deviation higher than the average proximity centrality of the group network, then the node... As a hub node; The set of nodes and pivotal nodes that exhibit significant differences is called the key brain region. For signals Screening key brain regions and extracting signals from corresponding channels , For the signal Key brain region subnetworks constructed.

4. The brain network evolution trajectory prediction method according to claim 1, characterized in that, The acquisition of the connection probability between nodes within a node includes: A brain functional network contains multiple node pairs; this refers to one of these node pairs. The connection probability between nodes within a node is expressed as: ； in: Represents a node and nodes Anatomical distance between; The constraint representing the anatomical distance is quantified using a power-law decay model. Represents a node and nodes Topological similarity between them; Based on the mutual information of multi-order paths between nodes within a node pair, the topological similarity between nodes within a node pair is obtained as the negative value of the conditional entropy, expressed as: in: Represents a node and nodes There are connections between them. node and nodes There is no connection between them; Represents a node and The length between is The set of all paths; This represents the maximum path length considered in a brain network; Indicates length is The weight of the path; ; , , Representing nodes respectively , and The degree.

5. The brain network evolution trajectory prediction method according to claim 1, characterized in that, The acquisition of the final simulated brain network includes: The evolution starts with the brain functional network of healthy individuals and ends with the brain functional network of patients. The edge difference rate between the number of edges in the brain functional network of healthy individuals and the number of edges in the brain functional network of patients is obtained, and edges are added or deleted based on the edge difference rate. Simulated brain networks during the iterative process The connection probabilities between any two nodes are sorted according to their magnitude. This represents the number of evolution iterations; right The edges in the network are updated. When an addition operation is performed, a specified connection is established between node pairs that do not have any connections, selected from the highest-ranked set of connection probabilities. When a deletion operation is performed, a specified connection is broken between node pairs that are known to have any connections, selected from the lowest-ranked set of minimum connection probabilities. This process yields the currently simulated brain network. .

6. The brain network evolution trajectory prediction method according to claim 5, characterized in that, The edge difference rate between the edges in the brain functional network of healthy individuals and the edges in the brain functional network of patients is expressed as follows: ； in: This represents the mean number of edges within the patient's brain functional network; Indicates the first The average number of edges in the simulated brain network obtained from each evolutionary iteration.

7. A brain network evolution trajectory prediction system, characterized in that, include: The information acquisition module is used to acquire brainwave signals from healthy individuals and patients with depression. Using the brain regions corresponding to each channel in the EEG signals of healthy individuals and patients with depression as nodes, and the brain regions transmitting functional EEG signals through the corresponding channels, a graph network is constructed to form a brain functional network that includes both healthy individuals and patients by using the connections between the brain regions corresponding to the channels as edges. Brain regions associated with depression were extracted from the brain functional network to form brain region subnetworks; in these subnetworks, corresponding nodes from the brain regions of healthy individuals and patients with depression were paired. The connectivity probability modeling module is used to obtain the anatomical distance between nodes within a pair of nodes in a brain functional network, and to impose an exponential decay constraint on the anatomical distance. Since there are multiple transmission paths between nodes within a pair to transfer information, the module obtains the mutual information of these multiple paths, and then obtains the topological similarity between nodes within the pair based on the mutual information. The product of the constrained anatomical distance and the topological similarity is set as the connectivity probability between nodes within the corresponding pair, and the module also obtains the connectivity probability set of all node pairs in the brain region subnetwork. The evolution module is used to set the evolution rules from the healthy brain functional network to the patient brain functional network, with the healthy brain functional network as the starting point and the patient brain functional network as the ending point. Based on the edge difference rate between the edges in the healthy brain functional network and the edges in the patient brain functional network, edges are added and deleted in the healthy brain functional network. When adding edges, the edge connection is established between the nodes with the highest connection probability in the connection probability set from the unconnected node pairs. When deleting edges, the edge connection is broken between the nodes with the lowest connection probability in the connection probability set from the connected node pairs. The brain functional network of healthy individuals is evolved into the brain functional network of patients according to the evolutionary rules, resulting in the final simulated brain network. The trajectory of the edges in the brain functional network of healthy individuals to the edges in the final simulated brain network is extracted from the evolutionary process, so as to predict the evolution trajectory of the brain network.

8. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the brain network evolution trajectory prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the steps of a brain network evolution trajectory prediction method as described in any one of claims 1 to 6.

Images